Evolving Lessons on the Complex Role of AMPK in Normal Physiology and Cancer

Abstract

AMP kinase (AMPK) is an evolutionarily conserved enzyme required for adaptive responses to various physiological and pathological conditions. AMPK executes numerous cellular functions, some of which are often perceived at odds with each other. While AMPK is essential for embryonic growth and development, its full impact in adult tissues is revealed under stressful situations that organisms face in the real world. Conflicting reports about its cellular functions, particularly in cancer, are intriguing and a growing number of AMPK activators are being developed to treat human diseases such as cancer and diabetes. Whether these drugs will have only context-specific benefits or detrimental effects in the treatment of human cancer will be a subject of intense research. Here we review the current state of AMPK research with an emphasis on cancer and discuss the yet unresolved context-dependent functions of AMPK in human cancer.

AMPK subunit structure: unanswered questions and the brewing controversy

Few other molecules in modern medicine have earned a status as controversial as AMP activated protein kinase (AMPK). The reasons, although less discussed, are apparent. AMPK is not a single-peptide enzyme; rather it is a complex of three subunits [1–3] that are chosen in a given cell from an available pool of seven subunits. Therefore, theoretically, twelve types of AMPK complexes can form in a cell, although specific complexes are preferentially formed in different tissues (see below). How each of the twelve possible trimeric complexes preferentially form and function under different contexts is mostly unknown. Little is known about whether each complex localizes to specific subcellular locale and serves distinct situation-dependent functions targeting specific set of substrates, and if the same AMPK complex functions differently in the diseased state compared to normal tissue. Therefore, a complex scenario has emerged where overly simplistic conclusions drawn from conflicting studies have in some cases caused significant confusion. In this review, we will briefly discuss AMPK structure and function in eukaryotes and elaborate on the current status of AMPK research in cancer and the complex issues that need to be addressed in the next few decades. A detailed description of the mechanisms of AMPK activation, AMPK structure and function in normal physiology and in other human diseases including diabetes, cardiovascular and neurological diseases can be found in several excellent reviews [1–5].

The AMPK heterotrimer: known and unknown

AMPK is a multisubstrate serine/threonine kinase. Purified AMPK was found to exist as a heterotrimer of catalytic α and regulatory β and γ subunits [6,7]. In mammals, there are two genes encoding the catalytic α1 and α2 subunits (called PRKAA1 and PRKAA2), two genes encoding the β1 and β2 subunits (PRKAB1 and PRKAB2), and three genes encoding the γ1, γ2 and γ3 subunits (PRKAG1, PRKAG2 and PRKAG3) [1–3]. The rooted trees of the α and β subunits suggest that vertebrate PRKAA1/A2 and PRKAB1/B2 genes arose by duplications of ancestral genes in lower eukaryotes. The evolutionary trees also show a higher mutation rate of β (1.8183 substitutions per site) compared to α (1.1024 substitutions per site) indicating that the β subunit evolved 1.65 times faster than the α subunit [8]. The N-terminus of the α subunit contains the catalytic domain as well as a phosphorylation site for upstream kinases that regulate its activity [9]. The γ subunits are nucleotide binding regulatory subunits that bind AMP/ADP. The conserved C-terminus of the β subunit interacts with both the α and γ subunits and plays an obligatory role in AMPK complex formation [10,11]. In addition, the β subunits contain a conserved carbohydrate binding domain that perhaps allows AMPK to function as a glycogen sensor [12,13].While the AMPK subunits are expressed more or less ubiquitously and the αβγ AMPK complex can come in twelve different flavors, the individual AMPK subunits in humans and mice display considerable variation in tissue-specific expression [14,15], subunit association, and subcellular localization.

Earlier studies demonstrated that the α2 catalytic subunit associates exclusively with β1 in slow twitch (type I) soleus muscle fibers, whereas in fast twitch (type II) extensor digitorium longus muscle fibers, α2 associates with both β1 and β2 subunits [16,17]. It is important to note that the preferred combination in a given tissue (or cell type within a tissue) is not static and can shift towards a different combination in response to a change in cellular physiology. The majority of these observations came from classic studies in skeletal muscle. For example, in human skeletal muscle three complexes are expressed in a preferred order (α2β2γ1>>α2β2γ3=α1β2γ1) and the preferred complex shifts depending on the intensity and duration of exercise [18–20]. The α2 AMPK complex while essential for basal AMPK function in skeletal muscle (and this function is not compensated by α1), is not required during adaptive response (e.g., exercise) during which the α1 AMPK complex can compensate for loss of α2 [21]. To add to this complexity is the recent finding that while the predominant AMPK complex in the human liver is α1βγ1, it is α1β1γ1 and α2β1γ1 in the dog and rat liver, respectively [22]. These interspecies differences could potentially complicate data interpretation when drawn from one species and applied to another. In mouse neural stem/progenitor cells, the AMPK complex containing the β1 subunit is predominant [10], yet the complex shifts to an equal β1 and β2-containing complex in differentiated neural cells like neurons and astrocytes (our unpublished data). Similar shift in AMPK subunit expression was also reported in myoblasts and myotubes [23] If cancer is a disease of undifferentiated stem/progenitor-like cells or de-differentiated cells, it is worth noting that in human glioblastoma cell lines the ratio of β1 and β2 is reminiscent of neural stem/progenitor cells [24]. Little is known about the regulation of AMPK subunits and the significance of AMPK complex switch during progenitor cell proliferation and differentiation. The subcellular localization of the AMPK subunits also varies. For example, in skeletal muscle, the α2, but not the α1 subunit translocates to the nucleus in response to exercise [25]. The β1 subunit is enriched in the nucleus of neurons [26] and neural stem cells [10], whereas β2 is predominantly cytoplasmic. The subcellular localization of β subunits is crucial for AMPK function, as studies in yeast have shown that they direct the localization of the catalytic α subunit and thus dictate the subset of substrates that can be phosphorylated in a given context [27]. In a recent study, compartmentalized AMPK signaling in subcellular regions has been demonstrated by the use of novel molecular sensors [28].

There are at least two other levels of complexity about AMPK heterotrimer formation and function that need to be addressed. First, we do not know much about the isoform (transcript variant) expression of the AMPK subunits in normal and diseased tissue. Our published [15] and unpublished results from germline β1 and β2 knockout mice revealed that at least three organs – the liver, kidney and skeletal muscle express multiple isoforms of β1 and β2 subunits. These isoforms were detected by protein analysis and these or additional isoforms may be expressed in other tissues at lower levels that remained undetected by our analysis. Whether AMPK isoforms are differentially expressed under various physiological or pathological conditions and whether this impacts AMPK complex structure, subcellular localization and substrate choice are unknown.

Second, results from our and other laboratories indicate that as yet unknown tissue-specific factors regulate AMPK complex formation and activity. This is because when compensatory increase in expression of one subunit occurs due to knockout of another subunit, AMPK complex formation and activity still remains uncompensated. For example, in the α2 knockout skeletal muscle, a significant compensatory increase of α1 protein occurs but AMPK activity remains downregulated [21, 29]. Similarly, in the β2 knockout muscle, the β1 subunit is upregulated several fold but AMPK activity remains ~70% below normal [15, 30]. Even the stability of the α subunits (that require β subunits for stability) remains affected in the presence of upregulated β1, suggesting unknown factors that regulate tissue-specific AMPK heterotrimer formation. These results tell us that much remains unknown about the tissue-specific regulation of AMPK subunits and activity in normal physiology and how AMPK complex formation, activity and substrate phosphorylation is impacted in a complex disease like cancer. Interpretation of the many conflicting results on AMPK function in healthy and diseased tissue may necessitate deeper understanding of these complex issues.

Is AMPK simply an energy Sensor?

AMPK regulation

AMPK was originally identified in a rat liver preparation as a contaminant of its now well-known substrate acetyl Co-A carboxylase (ACC), the rate limiting enzyme in fatty acid synthesis [31–33]. The enzyme was purified [34] and was found to greatly accelerate phosphorylation and inactivation of ACC in the presence of AMP, while reducing ACC phosphorylation to activate it at high ATP concentration. The enzyme was also found to inhibit HMG-CoA reductase, the rate limiting enzyme in cholesterol biosynthesis [35]. Therefore, this enzyme that was inhibited by ATP and activated by AMP and by phosphorylation, and was named AMP-activated protein kinase or AMPK [36]. It is important to note that AMPK is not the only enzyme regulated by ATP/AMP. Several metabolic enzymes (e.g., the glycolytic enzyme phosphofructokinase, the citric acid cycle enzyme citrate synthase and the glycogen metabolism enzyme glycogen phosphorylase) are sensitive to ATP/AMP ratio. However, a change in nucleotide ratio is probably one of the many ways by which AMPK is regulated (see Figure 1).

Figure 1. Modes of AMPK Activation.

Signaling through various cell surface receptors influence AMPK activity. In some contexts, AKT can inhibit AMPKα1 by phosphorylation; α2 is a poor AKT substrate. Much is unknown about the mechanisms by which AMPK is activated by these various upstream signals, particularly the AMPKK’s (LKB1 or CAMKKβ or TAK1) that mediate AMPK activation. The metabolic hormone leptin activates AMPK in the liver and muscle but inhibits hypothamic AMPK in the brain by unknown mechanisms. Metabolic stress activates AMPK by various mechanisms: some are mediated by LKB1; others go through CAMKKβ and in some case AMPK activation by one AMPKK may be intensified by another AMPKK. Some xenobiotics may diffuse through the plasma membrane bilayer while others may use channels and the mechanism by which the myriad xenobiotics activate AMPK is largely unknown. The question marks indicate pathways that need to be independently validated and require further investigation. PI3K: Phophatidylinositol-3-kinase; PLC: Phospholipase C; Gαq: G protein class αq that activates PLCβ; LKB1: Liver kinase B1; CAMKKβ: Calcium calmodulin kinase kinase β; TAK1: TGFβ activated kinase 1; ROS: reactive oxygen species; ER: endoplasmic reticulum.

To date, activation by AMP:ATP ratio and Ca⁺² are the best studied mechanisms of AMPK activation. When the cellular AMP:ATP ratio increases, AMP/ADP binds to the γ subunits. This interaction causes (a) allosteric activation, (b) a conformational change in the catalytic α subunits that enables activating phosphorylation at T172 in α1 and/or α2 by upstream kinases, and (c) inhibition of T172 dephosphorylation by phosphatases [37–39]. It was later found that binding of ADP can also activate AMPK [40–42]. There are at least three (probably more) kinases that can phosphorylate AMPK at T172 – Liver kinase B1 (LKB1) which is also known as Serine/threonine kinase 11 (STK11), Calcium/calmodulin-dependent kinase kinase (CaMKKβ) and TGFβ-activated kinase (TAK1) [43–49]. Of these LKB1 and CaMKKβ are the best studied, and they are the major upstream kinases for AMPK. LKB1 activates AMPK in response to AMP/ADP, while CaMKKβ activates AMPK in response to Ca⁺², and both pathways can act independently or synergistically [50]. G-protein coupled receptor (GPCR) activation and T cell receptor ligation to CD3 increase cellular Ca⁺² causing AMPK activation. Likewise, the calcium ionophore ionomycin also activates AMPK in a CAMKKβ-dependent manner [51]. The AMPK holoenzyme can also be activated by allosteric regulation of the β subunits with or without phosphorylation of T172 of the α subunits; however, this requires autophosphorylation at S108 in the β subunits (specifically the β1 subunit) [52]. AMPK activation without T172 phosphorylation was also observed in vivo [53, 54]. Therefore, for a better understanding of AMPK function in a given context, both T172 phosphorylation and phosphorylated levels of its substrate/s relevant to the specific context should be carefully considered.

What does AMPK sense?

It is important to note that AMPK senses cellular energy (and Ca⁺²) and not nutrient levels per se. One possible exception is glycogen that binds to the β subunits [55, 56], but the implications of such an interaction are not fully understood. AMPK has evolved to sense energy stress and to allow organisms to survive sudden periods of acute stress or longer lean periods when food is scarce. These episodes of environmental stress that organisms face in the wild are difficult, if not impossible, to recreate in the laboratory where animals are maintained in a constant environment, and therefore, the full gamut of AMPK function will likely remain unknown. Some examples include AMPK activation in overwintering/hibernating species, diapausing embryos, during drop in O₂ in the local environment (estivating fish) and acute hypoxia [57–64]. It is important to note that AMPK is also a sharp sensor of cellular osmotic pressure and cellular entry of a host of xenobiotics that includes synthetic small molecules, drugs, pro-drugs, and natural products [1, 65, 66].

There are reports that some xenobiotics either inhibited mitochondrial function (thus possibly increased the AMP:ATP ratio) or elevated intracellular Ca⁺² [65], and these were proposed as the mechanisms of AMPK activation by xenobiotics. One should note that multiple studies including ours [22, 67] found that the antidiabetic drug and mitochondrial Complex I inhibitor metformin (an indirect AMPK activator) did not change nucleotide ratio, but still activated AMPK, and in contrast to the observation of Hawley et al., (65) in HEK293T cells, we found no change in AMP:ATP ratio in resveratrol-induced AMPK activation in Neuro2A cells [68]. While most cancer cells efficiently compensate for mitochondrial ATP loss by augmenting glycolysis as we found in glioma cells [24] which may mask any chronic change in nucleotide ratio, other mechanisms including an AMP/ADP-independent function of AMPK as a sensor of cellular insult or simply an altered physiological state [62] cannot be ruled out. Xenobiotic sensing occurs through the P450 system that includes several genes transcribed by the nuclear receptors CAR and PXR [69]. AMPK was found to be necessary for xenobiotics-induced transcription [70–72] and an important unanswered question is whether xenobiotic-sensing of AMPK is an evolutionarily conserved function that occurs through AMPK-dependent CAR and PXR regulation, independent of its energy or Ca⁺² sensing functions. It is noteworthy that cellular pathology (e.g., cancer) exhibits various stresses including oncogenic, nutrient, energy, O₂, osmotic and shear stress along with chemotherapy and radiotherapy-induced stress [73–77], and it is possible that active AMPK provides stress-resistance in human cancers. Physiological, pathological and drug-induced AMPK activation likely occurs through different mechanisms with unique context-dependent consequences and a crucial question that warrants investigation is the role of different AMPK complexes that contribute to AMPK activity under these myriad contexts.

The Many functions of AMPK

AMPK function in normal growth and development

The AMPK heterotrimer is ubiquitously expressed. The canonical function of AMPK is to inhibit anabolism and activate catabolism (see Figure 2). Embryonic lethality of AMPK null lower eukaryotes and mice [11, 78–81] suggests that AMPK is indispensable for early development. Genetic studies in model organisms confirmed that loss of AMPK causes various metabolic phenotypes in mice [15, 27, 81, 82], and defects in cell polarity, growth, development and longevity of both plants and animals [11, 78, 79, 81, 83–87]. However, the most enigmatic functions of AMPK are in mitosis and cell proliferation. Active AMPK localizes to the mitotic spindle of mammalian cells [88] and AMPK was shown to phosphorylate 28 novel proteins involved in chromosomal segregation, mitosis, cytokinesis and cytoskeletal reorganization [89]. The physiological implications of these observations are unclear and warrant further investigation. The critical requirement of AMPK for normal cell cycle progression is also found in mammalian and Drosophila cells [78, 90–92] and in a whole genome RNAi library screen of mammalian cells [93]. We earlier discovered that AMPK phosphorylates the retinoblastoma (RB) protein to allow cell cycle progression of neural progenitor cells [10] and a strong correlation between AMPK and RB phosphorylation was recently confirmed by another group [94]. In an unbiased proteomic screen of phospho-proteins enriched in mitosis, high levels of phosphorylated AMPK α2 was found, and importantly, this phosphorylation was maintained at high ATP levels [95]. Whether AMPK’s role in mitosis is uncoupled from or connected to its energy sensing function is unclear. A low energy level (when AMPK is known to be activated) is understandably a ‘Stop’ signal for cells to enter energy-intensive cell cycle. Yet, as noted above, active AMPK is found in dividing cells.

Figure 2. AMPK Downstream Targets.

AMPK directly or indirectly phosphorylates multiple substrates to increase glucose transporter translocation from cytoplasmic vesicles to plasma membrane and promote glycolysis. Through inhibitory phosphorylation, AMPK is a negative regulator of lipogenesis (green lines) and inhibits mTORC1 by phosphorylating (activating) TSC2 or inhibiting RAPTOR, the mTORC1 binding partner. Both mTOR and AMPK phosphorylates ULK1 at different sites to inhibit and activate autophagy, respectively. The question marks indicate substrates/pathways that require independent validation and further investigation. ACC: Acetyl-CoA carboxylase; HMG-CoA: 3-hydroxy-3-methylglutaryl CoA; SREBP1: Sterol regulatory element-binding protein 1; HSL: Hormone sensitive lipase; TSC2: Tuberous Sclerosis Complex 2; ULK1: Unc-51-like kinase 1; eNOS: endothelial nitric oxide synthase; CREB: cAMP response element-binding protein; H2B: Histone 2B; MLC2: Myosin light chain 2; CLIP170: Cytoplasmic linker protein 170; MLCK: Non-myosin light chain kinase; PAK2: p21-activated kinase 2; PPP1R12C: Protein phosphatase 1, regulatory subunit 12C; RB: Retinoblastoma; PGC1α: Peroxisome proliferator-activated receptor gamma coactivator 1-α; HIF1α: Hypoxia-inducible factor 1α; PFK2: Phosphofructokinase 2; PFKBP3: 6-Phospho-2-kinase/fructose-2–6 bisphosphatase; GS: Glycogen synthase; TBC1D1: Tre-2/USP6, BUB2, cdc16 domain family member 1; MEF2: Myocyte enhancer factor 2; GEF: GLUT4 enhancer factor.

How can we reconcile these novel findings with the canonical role of AMPK as a low energy sensor? A decision to enter cell cycle is dictated by growth factor signaling, and we and others have observed a correlation between growth factor stimulation and AMPK activation that occurs possibly through the interaction of AMPK with growth factor adaptor proteins such as Src and Grb2 [10,96–101]. Should these observations be considered outliers in the backdrop of the overwhelming literature about AMPK’s role in growth suppression? Could it be that specific subcellular pools of AMPK heterotrimers are activated in an energy-rich context, independent of AMP/ATP ratio to coordinate proper cell division, while other AMPK complexes simultaneously perform the canonical functions of AMPK? Further research is needed to explain these seemingly discordant roles of AMPK in normal physiology.

The controversy about AMPK function in cancer

It is fair to state that there is no single unifying role of AMPK in cancer. The debate as to whether AMPK promotes or suppresses tumor growth stems from (a) the use of non-specific and promiscuous pharmacological agents that inhibit or indirectly activate AMPK and therefore create a discrepancy between pharmacological and genetic studies, (b) our limited understanding of context-dependent and cell type-dependent AMPK heterotrimer composition, subcellular localization and substrate selectivity, (c) oversimplification of AMPK-regulated pathways, which often fails to discriminate between signaling in normal cells and anomalous signaling in cancer cells, and (d) the absence of genetic models to rigorously examine the role of AMPK role in different stages of tumor development. The most prominent of the pharmacological agents in this context are the anti-diabetic drug metformin, the AMP mimetic AICAR and the multi-kinase inhibitor Compound C. A cause and effect relationship was initially drawn between AMPK activation by AICAR, metformin (and other synthetic compounds and natural products) and inhibition of cancer cell proliferation [102–107]. We and others have now convincingly demonstrated that AICAR, metformin and Compound C [22,108–111] are highly non-specific, and all three agents inhibit proliferation through AMPK-independent mechanisms. AMPK activation by xenobiotics per se is not a cause of cell death; instead AMPK activation occurs as a response mechanism to protect cells from the toxicity of these agents. This proposal is consistent with the increased vulnerability of AMPK null hepatocytes [112], LKB1 mutant lung cancer cells [113] and AMPK null glioma and stromal cells [24] to AICAR or metformin.

Contrary to the above pharmacological studies, experiments in human cancer cells using genetic tools demonstrated that AMPK supports cancer cell survival, proliferation and migration through redox homeostasis, by cooperating with oncogenes (Myc), augmenting cellular transformation and metabolic reprograming, regulating microtubule dynamics, providing protection from chemotherapy and radiation, and through unknown mechanisms [114–128]. The caveat is that most of these are in vitro studies that need to be tested in vivo. A few in vivo studies that indicate that AMPK has a contrasting role in cancer includes its tumor promoting role in RAS-transformed fibroblasts, astrocytic and breast tumors [119,120,125], and its tumor suppressive role in a mouse model of Myc-induced lymphoma [129]. Recently, in vitro genetic studies also showed a growth suppressive action of AMPK through its regulation of the Hippo pathway [130–132]. Examining the functions of a stress sensor such as AMPK in cell lines adapted in culture for decades that are grown under the artificial and most favorable cell culture conditions (e.g., 25 mM glucose and ambient air) is not the best approach. For mechanistic studies, all attempts should be made to include primary cancer lines freshly established from patient tumors that have not been passaged for years. Another option for in vitro mechanistic studies would be organoid cultures or other 3-D culture systems maintained at physiological nutrient and oxygen levels – these systems could potentially allow formation of nutrient and oxygen gradients mimicking solid tumors and would be clinically more relevant for studying functions of stress sensors like AMPK. However, although significantly more challenging for mechanistic studies, the true functions of AMPK in cancer initiation, promotion or suppression will be revealed only by mouse models and genetic patient-derived xenograft (PDX) models.

The AMPK pathway in the human cancer genome

The rationale that supports a tumor suppressive role for AMPK was based on the discovery of the tumor suppressor LKB1 (STK11) as the upstream AMPKK that activates AMPK [44]. Germline mutations in LKB1 causes benign hamartomas in the Peutz-Jeghers cancer syndrome (PJS), and PJS patients have a higher risk of developing mainly intestinal and breast cancer [133–135]. A distinction should however be drawn between LKB1 and other tumor suppressors (TS) that are frequently inactivated in human cancer. The mutational landscape of major human cancers revealed that unlike the two frequently mutated TS genes - TP53 (mutation frequency range: 2–94%; pan-cancer frequency: 42%) and PTEN (mutation frequency range: 2–63%, 94%; pan-cancer frequency: 9.7%), pan-cancer frequency of LKB1 mutation is one of the lowest among TS genes (0.9%) and except for lung adenocarcinoma and uterine cancer, where LKB1 mutation frequency is around 10% and 1.7%, respectively, LKB1 is rarely mutated in human cancer (mutation frequency range: 0–0.4%) [136–138]. One can argue that in the tug of war between the throttle and gas pedal (keeping LKB1 may slow down proliferation but increase survival in the tumor milieu), most tumors may have evolved to retain rather than mutate LKB1. Indeed, contrary to other TS genes that suppress transformation, both LKB1 and AMPK are required for oncogene-driven transformation of fibroblasts, and LKB1 null tumor cells are preferentially killed by stress-inducing drugs [44,113,119].

Two other important aspects about the LKB1-AMPK axis in cancer that need to be highlighted is that first, LKB1 has at least 13 targets other than AMPK [139] and some of LKB1’s effects are likely mediated by substrates other than AMPK. Indeed, expression of a hypomorphic LKB1 allele in mouse caused AMPK-dependent branching defects of the developing lung, while pancreatic cyst formation was AMPK-independent [140]. Second, the other major AMPKK, CAMKKβ is not a TS gene in human cancer [136–138]. In fact, CAMKKβ is overexpressed up to 7-fold in 92/98 (94%) of gastric cancer [141], and played an AMPK-mediated role in metabolic reprograming, migration and invasion in prostate and colon cancer [126,128,142,143]. It is reasonable to predict that oncogene-induced ER stress and calcium disequilibrium can activate a specific pool of the AMPK heterotrimer through CAMKKβ.

Analysis of AMPK mutation data in human cancer [144,145] shows interesting results. The AMPK α2 subunit is mutated at a frequency of 0.2 – 10% across all human cancers (highest in small cell lung cancer and melanoma, but under 4% in all other cancers). The only cancer where α2 is amplified is sarcoma (~6%). On the contrary, the AMPK α1 subunit is rarely mutated but instead amplified at a frequency of 1 – 15% (15% in lung squamous cell and adenocarcinoma, and between 5–10% in sarcoma and cancers of the stomach, bladder, esophagus, head and neck, uterus and ovary). Is the AMPKα1 complex oncogenic and the AMPKα2 complex growth suppressive in a context-dependent manner? Supporting the above argument, deletion of α2 enhanced while deletion of α1 suppressed growth of RAS-transformed murine fibroblasts [146]. However, opposing the argument are reports from in vitro studies that AKT activation specifically reduces AMPKα1 activity by phosphorylating α1 at S487, while having no effect on α2 which is a poor ATK substrate [147–149]. The significance of these findings remains to be tested in vivo. The frequency of β1 subunit mutation is low (<2%), except nerve sheath tumors where it is deleted (~13%). However, it is amplified in sarcoma, prostate, ovarian and stomach cancer (3–7%). In sharp contrast, the β2 subunit is amplified across nearly all cancers (4–16%). Among the γ subunits, the only cancer where γ1 is significantly deleted (16%) is nerve sheath tumors, while the only cancers where data on the γ2 subunit are consistent are ovarian cancer, where it is amplified (~12%), and AML where it is deleted in about 4% of cases. Remarkably, despite overexpression in gastric cancer, CAMKKβ is deleted in about 13% of nerve sheath tumors and ~4% of melanoma. Experimental validation studies will be required to test the functional impact of these mutations. Besides these genetic changes, epigenetic mechanisms significantly influence gene expression and much work is needed to understand the regulation of AMPK subunits, LKB1 and CAMKKβ in human cancer.

The AMPK - lipogenesis axis in cancer

AMPK phosphorylates ACC to inhibit its activity (Figure 2). A distinction between the consequences of inhibition of ACCA versus ACCB that is often overlooked and requires particular attention in cancer is that ACCA inhibition blocks de novo fatty acid (FA) synthesis, while ACCB inhibition by AMPK increases FA oxidation [150]. One overwhelming argument is that AMPK activation would inhibit ACCA, block lipogenesis and reduce tumor growth. Indeed, the high expression of FASN (the FA synthesizing gene) is correlated with poor prognosis in many cancers, and FASN inhibition caused apoptosis in vitro and in vivo [151]. However, it remains unclear if the main cause of cancer cell death is inhibition of FA synthesis per se or cytotoxic effects of accumulated malonyl-Co-A that occurs after FASN inhibition [152]. This is an important point to ponder because FASN or ACCA knockdown does not kill lung, glioma or breast cancer cells [24,114,153]. In fact, ACCA or ACCB knockdown or inhibition by AMPK is required to maintain NADPH levels and in vivo growth of lung and breast tumor xenografts [114]. We reported that expression of an ACCA mutant that cannot be phosphorylated by AMPK had no effect on glioma cell proliferation. However, reducing exogenous lipids alone inhibited proliferation [24] suggesting that dietary lipids could be crucial for certain cancers.

Given that the flow of glucose to the FA precursor citrate is limited due to the Warburg effect (preferential glycolytic metabolism in the presence of oxygen), and citrate production by reductive decarboxylation of glutamine-derived α ketoglutarate may not be sufficient to compensate for the Warburg effect, some cancers may partially uncouple the negative regulation of AMPK on ACCA (and thus de novo lipogenesis), and become more reliant on dietary lipids. Recent studies underscore the importance of macromolecule scavenging by cancer cells [154,155] and dietary FAs could be scavenged or transported through FA transporters to support tumor growth. Interestingly, FASN and ACCA are not greatly amplified in human cancer (range – 3–9% and 0.5–5, respectively), while ACCB is mutated in a wide range of cancers (2–13%), and at least some genes in the FA oxidation and transport pathway (CPT1A, SLC27A1, SLC27A3) are amplified in some cancers (range 3–30%; 2–12% and 2–15%, respectively) [141,142]. To target de novo lipogenesis, ACCA inhibition by specific inhibitors rather than by AMPK activation should be a better strategy to avoid unintended consequences of AMPK activation like tumor survival through autophagy, enhanced glycolysis, macromolecule import and redox homeostasis [111,122,156–158] and other pro-tumor functions of AMPK.

The AMPK - mTOR axis in cancer

AMPK phosphorylates the tumor suppressor TSC2 and the mTORC1 partner RAPTOR to inhibit mTOR (Figure 2) [159–160]. While there is no question this regulatory mechanism exists, an oversimplified linear view of this pathway together with drawing a parallelism between normal and cancer cells has led to apparently incompatible results from different laboratories. Although AMPK activation during glucose stress represses mTOR in vitro to increase cell survival in some but not all cases [114,161], we found that AMPK silencing in glioma cells did not do the opposite, i.e., activate mTOR [24], suggesting either the existence of other ways of controlling mTOR or interference between AMPK and mTOR due to the complex rewiring of signaling networks in cancer cells. Indeed, mTOR regulation is highly context-specific and can be controlled independent of AMPK. For example, we earlier reported that in skeletal muscle, AMPKβ2 knockout caused an expected basal mTOR activation that was greatly amplified during exercise when AMPK activation suppressed mTOR in wildtype mice [15]. However, nutrient stress in vivo (fasting) inhibited mTOR independent of AMPK [15]. In fact, mTOR inhibition can occur through REDD1 and p38 independent of TSC2 in a context-dependent manner [162–164]. Therefore, AMPK is not absolutely necessary for mTOR inhibition suggesting that in certain contexts, activated AMPK and mTOR may co-exist. Indeed, this is what we observed in glioblastoma suggesting that basally active AMPK (and possibly only specific AMPK heterotrimers) may not be sufficient to put a brake on mTOR [24]However, in-depth analysis of tumor tissue by immunohistochemistry and flow cytometry is required to examine if the two signals exist in same or distinct cells.

Global activation of alternative splicing, including aberrant splicing of genes, occurs in human cancers in response to oncogene activation, tumor suppressor loss, tumor hypoxia and other changes in the tumor microenvironment [165–167]. Many of these splice variants play roles during development but are reactivated in cancer to evade growth suppressors, apoptosis, immune detection and to engage in tumor metabolic reprograming. In prostate cancer, androgens drive alternative splicing of TSC2, resulting in the dominant expression of a splice variant that lacks TSC1 and RHEB interacting domains leading to constitutive mTOR activation [168] and potentially de-linking AMPK and mTORC1. If proven true in other contexts and tumors, this is a fascinating mechanism of “having the cake and eating it too” that may allow simultaneous AMPK and mTOR activation.

A third and potentially interesting mechanism by which cancer cells may evade AMPK’s inhibition of mTOR-regulated protein translation is by switching from CAP to IRES (Internal Ribosomal Entry Site) - dependent translation that often occurs during oncogenesis [169–171]. AMPK is activated while mTOR is inhibited during nutrient stress and hypoxia that are common features in advanced tumors [172,173]. In stressed areas of tumors, a switch to IRES-mediated, mTOR-independent translation allows synthesis of crucial proteins involved in growth, proliferation, angiogenesis, survival, and resistance to chemotherapy and radiation. Some examples include cMyc, VEGF, XIAP, APAF1, FGF2, TrkB and Aurora kinase [174–177]. Therefore, it is conceivable that in a heterogeneous tumor environment, high AMPK and high mTOR signaling can co-exist in regions of high and low stress, respectively, and AMPK and mTOR signaling could be in dynamic flux between the different cell populations. Indeed, coordinated regulation of AMPK and mTOR signaling occurs during human mesenchymal stem cell differentiation [178]. An interesting question is whether certain AMPK subunits are also translated by the IRES-mediated mechanism in high stress areas of solid tumors.

Concluding Remarks

A renewed interest in examining mitochondrial function in human cancer has led to the discovery of multiple aspects of tumorigenesis that are regulated by mitochondria. These include metabolic reprograming like reductive carboxylation of α-ketoglutarate, ROS production, metastasis, epithelial-mesenchymal transition (EMT) and drug resistance among others [179–185]. AMPK plays an important role in promoting mitochondrial biogenesis and function in normal tissue and cancer [116,126,186,187]. Therefore, mitochondria-dependent functions during tumor evolution may be favored by active AMPK in certain tumors. A hallmark of cancer metabolic reprograming is aerobic glycolysis (Warburg effect). Several studies now show that active AMPK supports glycolysis in human tumors [116,125,156,188,189], although it countered the Warburg effect in a mouse model of lymphoma [129], reflecting a possible tumor or species-specific role of AMPK in glycolysis. Is AMPK activated or inhibited during transformation to cope with transformation-associated metabolic reprograming? It is unclear whether all cells in a solid tumor switch to glycolytic metabolism or if genetic and intratumoral environmental heterogeneity require some cells to rely more on mitochondrial oxidation of glucose, fatty acids and other substrates (see Outstanding Questions).

Cells with a proliferative advantage in nutrient-rich environments may initially suppress AMPK to boost biosynthetic processes. Indeed, the AMPKα2 transcript is suppressed in Stages I &II but upregulated in grades III and IV of human gastric cancer [190]. As tumor cells undergo clonal expansion with accumulating mutations, and the tumor grows in volume, oncogenic stress, fluctuating nutrient and oxygen levels, malformed neovasculature and harsh acidic environment may force tumor cells to depend on AMPK and other stress enzymes to support their altered metabolism and survival (Figure 3). When AMPK suppresses proliferation by inhibiting mTOR while supports tumor energy balance and reprogramed metabolism, what is the rationale to activate AMPK to inhibit mTOR when specific mTOR inhibitors are available? A similar logic applies to inhibition of lipogenesis in cancer. Considering the potential risk of enhancing the diverse pro-tumor effects of AMPK, including the risk of relieving S6K suppression of IRS1 and mTORC2 activation during mTORC1 inhibition [191,192] by AMPK activators, a combined inhibition of mTOR and AMPK may prove to be a more effective strategy for certain advanced tumors. However, in cancers in which the net role effect of AMPK is growth suppression, specific and direct AMPK activators could be therapeutically more beneficial. In this regard, in a recent report, AMPK activation by relatively specific AMPK activators killed AML cells only with hyperactivated mTORC1 [193]. Interestingly, similar to our previous results [26], this study also found a disconnect between AMPK activation and mTOR suppression since AMPK activation did not inhibit mTORC1 in at least three AML cell lines and primary AML samples. With the caveat that human tumors are distinct from mouse tumors in many aspects, rigorous and parallel studies in genetically engineered mouse models and PDX models using controlled and tunable AMPK expression and inhibition will reveal the tissue specific effect of AMPK on tumorigenesis and progression.

Figure 3. Context-dependent role of AMPK in Cancer.

The cartoon shows probable roles of AMPK during the evolution of solid tumors. On one hand, specific genetic changes in cells may up-or downregulate AMPK activity, while on the other hand, the consequence of AMPK up-or downregulation may be diverse in cells of different genetic backgrounds and tissues. The cartoon is just one example in which loss of function (LOF), or gain of function (GOF) of tumor suppressors (TS) and oncogenes, respectively, may initially downregulate cytoplasmic AMPK activity to relieve its control on mTOR rendering a proliferative advantage to cells. Simultaneously, an active nuclear pool of AMPK may survey and ensure proper division of transformed cells. As tumor cells multiply, individual cells accumulate additional stochastic mutations and epigenetic changes leading to rise of clones, clonal competition, expansion and tumor heterogeneity. Each of these clones continues to evolve as the tumor itself evolves into an organized structure. These clonal subsets may conform to the principle of ‘division of labor’ such that subsets with high levels of either AMPK or mTOR may serve different functions and support each other to sustain overall tumor growth. In these subsets of cells, mTOR and AMPK signature may not be static but could be could be dynamically regulated. Active AMPK likely mitigates oncogenic stress and regulates metabolic reprograming. Finally, the hostile environment of solid tumors, particularly the specific subdomains of overgrowth, poor supply of nutrients and O₂, exert considerable strain on tumor growth. In these regions, AMPK may allow metabolic plasticity and survival and potentially metastasis by regulating anchorage-independent survival and migration through blood vessels and normal parenchyma.

TRENDS BOX.

• AMP activated protein kinase (AMPK) is an evolutionarily conserved cellular energy sensor

• There are many recognized and unrecognized roles of AMPK in normal and pathophysiology

• Some functions of AMPK are likely beyond energy sensing

• The multisubunit structure and sensitivity to various stresses confer complexity to AMPK function

• Appropriate genetic models are desperately needed to understand the context-dependent role of AMPK in metabolic diseases including cancer

Outstanding Questions.

• How do individual AMPK complexes preferentially form under different contexts?

• Are there complex-specific and context-specific preferred AMPK substrates?

• Can we conclude on the net effect of AMPK activation from a snap shot of phospho-signal on only one of its substrates (e.g., ACC)?

• Can the same complex function differently in diseased versus normal cells?

• Does the myriad stimuli that activate AMPK (physiological, pathological and drugs) require different AMPK regulatory (β/γ) subunits?

• How is tissue-specific expression of AMPK subunits regulated at the genetic and epigenetic level? Are there tissue-specific factors that allow formation of certain AMPK complexes only?

• Are AMPK transcript variants differentially expressed under normal physiology versus pathology and do they regulate AMPK biology?

• How are the AMPK subunits regulated during development, and what is the significance of subunit switch?

• Is xenobiotic sensing a conserved function of AMPK that occurs through mechanisms independent of energy or Ca⁺² sensing?

• Is AMPK signaling binary/digital (on or off) or analog (continuous gradient)? Are there ‘better’ AMPK substrates (that can get phosphorylated at low AMPK activity), and ‘poor’ AMPK substrates?

• Does AMPK mediate spindle formation, chromosomal segregation, cytokinesis and cytoskeletal reorganization? Are AMPK null cells error-prone during DNA replication and susceptible to aneuploidy? Is AMPK’s role in mitosis uncoupled from its energy sensing function? Could it be that specific subcellular pools of AMPK heterotrimers are activated in an energy-rich context, independent of AMP/ATP ratio to coordinate proper cell division?

• Is AMPK activated or inhibited during transformation to cope with transformation-associated metabolic reprograming? What factors dictate tissue or even species-specific role of AMPK in tumor suppression or promotion?